I hadn't thought of it that way before, nice point. I guess you could look at a complex adaptive system as a learning system that gradually discovers the best way to exploit its environment. Over-learning is what leads to the release phase (when the environment goes outside of your expected bounds and causes problems),and our meddling to induce extra randomness serves the same purpose here as it does in AI systems - ensuring that all of the variation of reality is represented, and a bit more.